Discovering and Visualizing Tactics in a Table Tennis Game Based on Subgroup Discovery

Abstract

We report preliminary results to automatically identify effective tactics of elite table tennis players. We define these tactics as subgroups of winning strokes that table tennis experts seek to identify in order to train players and adapt their strategy during play. We first report how we identify and classify these subgroups using the weighted relative accuracy measure (WRAcc). We then present the subgroups using visualizations to communicate these results to our expert. These exchanges allow rapid feedback on our results and makes it possible further improvements to our discoveries.

Exploring Table Tennis Analytics: Domination, Expected Score and Shot Diversity

“…In a previous paper [3], we saw that some players tend to serve in the same way, while they did not lose a point after such a serve. Here, we are going further in the sense that we explore more strokes into the rally, and because we create a metric representing the distance between two openings.…”

“…Meanwhile, a plethora of advanced tools are beginning to leverage this data, such as iTTvis [13], which is aimed at experts to explore game sequences and discover tactics. Other approaches also focus on sequence analysis to visually explore frequent patterns [3], using an a-cyclic graph to represent all points in a match, and extract tactics. TIVEE [2] leverages shot types, player positions, and shuttle trajectory and speed to find correlations between strokes, aiding in the discovery of tactics.…”

“…Fig. 1: Example of an extended table tennis detailed data model (from [3]). It includes additional metadata (e. g., players' names, score and winner) with advanced strokes types, players and ball rebound zones.…”

“…Such data can be collected with regular technical skills using a blend of computer vision, deep learning and manual annotation tools. It extends the format previously used in [3] but with finer grained players position, orientation and ball rebound position. Some researches have been led to collect such data automatically.…”